Download Assessing the applicability of quantum corrections to classical

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Double-slit experiment wikipedia , lookup

Quantum electrodynamics wikipedia , lookup

Renormalization group wikipedia , lookup

Relativistic quantum mechanics wikipedia , lookup

Scalar field theory wikipedia , lookup

Renormalization wikipedia , lookup

Max Born wikipedia , lookup

Bell's theorem wikipedia , lookup

Quantum field theory wikipedia , lookup

Wave–particle duality wikipedia , lookup

Density matrix wikipedia , lookup

Quantum entanglement wikipedia , lookup

Copenhagen interpretation wikipedia , lookup

Particle in a box wikipedia , lookup

Quantum dot wikipedia , lookup

Many-worlds interpretation wikipedia , lookup

Path integral formulation wikipedia , lookup

Hydrogen atom wikipedia , lookup

Quantum fiction wikipedia , lookup

Orchestrated objective reduction wikipedia , lookup

Theoretical and experimental justification for the Schrödinger equation wikipedia , lookup

Symmetry in quantum mechanics wikipedia , lookup

Coherent states wikipedia , lookup

EPR paradox wikipedia , lookup

Quantum computing wikipedia , lookup

Interpretations of quantum mechanics wikipedia , lookup

History of quantum field theory wikipedia , lookup

Quantum group wikipedia , lookup

Quantum teleportation wikipedia , lookup

Quantum key distribution wikipedia , lookup

Quantum machine learning wikipedia , lookup

Quantum state wikipedia , lookup

Hidden variable theory wikipedia , lookup

T-symmetry wikipedia , lookup

Canonical quantization wikipedia , lookup

Transcript
Assessing the applicability of quantum corrections to classical
thermal conductivity predictions
J. E. Turney,1 A. J. H. McGaughey,1, ∗ and C. H. Amon1, 2
1
Department of Mechanical Engineering
Carnegie Mellon University
Pittsburgh, PA 15213
2
Department of Mechanical and Industrial Engineering
University of Toronto
Toronto, Ontario M5S 3G8
(Dated: May 29, 2009)
Abstract
The validity of the commonly-used quantum corrections for mapping a classically-predicted thermal conductivity onto a corresponding quantum value are assessed by self-consistently predicting
the classical and quantum thermal conductivities of a crystalline silicon system via lattice dynamics calculations. Applying the quantum corrections to the classical predictions, with or without
the zero-point energy, does not bring them into better agreement with the quantum predictions
compared to the uncorrected classical values above temperatures of 200 K. By examining the
mode-dependence of the phonon properties, we demonstrate that thermal conductivity cannot be
quantum-corrected on a system level. We explore the source of the differences in the quantum and
classical phonon relaxation times on a mode-by-mode-basis.
1
I.
INTRODUCTION
Classical molecular dynamics (MD) simulation is a powerful tool for predicting physical
behavior and material properties.1–3 As a general rule, MD simulations of solids are considered valid near and above a material’s Debye temperature, where all of the vibrational
modes (i.e., phonons in a crystal) are fully excited. Molecular dynamics simulations are
not valid at lower temperatures, where quantum effects cannot be neglected. To mitigate
this limitation, quantum corrections (QCs) have been developed. The purpose of QCs is
to map predictions made for a classical system onto corresponding values in a quantum
system. Quantum corrections have been derived from first principles for thermodynamic,
mechanical, and structural properties (e.g., Helmholtz free energy, entropy, bulk modulus,
and pair distribution function).4–6 Rigorous QCs for transport properties such as thermal
conductivity are more difficult to obtain.7
The commonly-used QCs for thermal conductivity are based on ad-hoc physical arguments rather than fundamental theory. These QCs are applied to the temperature, T , and
the thermal conductivity, k.8,9 The temperature correction is made by equating the total
energies of the classical and quantum systems. The energy equality can be expressed as
N,3n
X
N,3n
E C(κν) =
κ ,ν
X
E Q(κν) ,
(1)
κ ,ν
where the sums are over all phonon modes denoted by N wave vectors, κ , and 3n dispersion
branches, ν, E(κν) is the energy of a phonon mode, and the superscripts C and Q indicate
classical and quantum. By assuming (i) equipartition of the classical energy and subtracting
the three translational degrees of freedom and (ii) the quantum energy to be harmonic, Eq.
(1) becomes
N,3n ·
C
3(N n − 1)kB T =
X
~ω(κν) f Q(κν)
κ ,ν
¸
~ω(κν)
.
+
2
(2)
Here, kB is the Boltzmann constant, ~ is the Planck constant divided by 2π, and ω(κν) and f(κν)
are the mode-dependent frequency and equilibrium occupation number. For the quantum
system, the equilibrium occupation number follows the Bose-Einstein distribution,
fQ =
³
exp
2
1
~ω
kB T Q
´
−1
.
(3)
The term ~ω(κν) /2 in Eq. (2) is the zero-point (ZP) energy. The ZP energy has no parallel
in a classical system, and, as will be shown, plays an important role in the QCs. Equation
(2) cannot be explicitly solved for T Q , but T Q can be determined by computing T C for a
series of T Q values and then interpolating.
The thermal conductivity correction is
kQ = kC
dT C
,
dT Q
(4)
which arises from equating the heat fluxes obtained from the Fourier law in the classical and
quantum systems. The QCs given by Eqs. (2) and (4) have been applied to classical MD
thermal conductivity predictions for amorphous silicon,9 crystalline silicon in bulk10–12 and
thin films,13 β-silicon carbide,14 silicon-germanium superlattices,15 carbon nanotubes,16 and
transition metals (the phonon contribution).17
The validity of Eqs. (2) and (4) is questionable as they are not derived from fundamental
relations and have not been rigorously tested. The doubts regarding these QCs are further
exemplified by the non-consensus on whether to include the ZP energy in Eq. (2) or to
ignore it.18 The objective of this work is to rigorously assess the validity of the QCs for
thermal conductivity given by Eqs. (2) and (4). To do so, we self-consistently predict the
classical and quantum thermal conductivities of crystalline silicon via lattice dynamics (LD)
calculations, and then apply the QCs to the classical predictions. We find that the QCs are
not valid, and demonstrate why using phonon properties obtained from the LD calculations.
II.
LATTICE DYNAMICS METHOD FOR PREDICTING PHONON PROPER-
TIES
We predict the classical and quantum thermal conductivities of an ordered (i.e., crystalline) material in a self-consistent manner using LD calculations. Consider a crystal with
N lattice sites and an n-atom unit cell with periodic boundary conditions applied in all directions. Harmonic LD is first used to find the phonon frequencies and polarization vectors,
e, (i.e., the normal modes of vibration) under the harmonic approximation by solving the
eigenvalue problem
κ) e(κν) .
ω 2(κν) e(κν) = D(κ
3
(5)
κ), has dimension 3n × 3n and elements
The dynamical matrix, D(κ
¯
N
¯
n h ¡ 0 ¢ ¡ ¢io
X
∂ 2Φ
1
¯
κ· r bl 0 −r 0b
κ) = √
,
D3(b−1)+α,3(b0 −1)+α0(κ
¡ l0 ¢ ¯ exp iκ
¡0¢
mb mb0 l0 ∂rα ∂rα0 0 ¯
b
b
o
(6)
where b and α index over the atoms in the unit cell and the Cartesian coordinates, Φ is the
¡¢
total potential energy, r bl is the average position vector for the bth atom in the lth unit cell
¡¢
with rα bl its α-component, and m is the atomic mass.19,20
The harmonic phonon frequencies and polarization vectors are then used as input to an
anharmonic LD calculation to compute the mode-dependent phonon frequency shifts, ∆,
and linewidths, Γ. For a quantum system these two quantities are determined for mode 0
(i.e., κ0 and ν0 ) from21,22
∆Q
0 =
N,3n N,3n
i
~ X X κ 0 κ 1 κ 2 2nh Q Q ih
−1
−1
|Φ(ν0 ν1 ν2)| f1 +f2 +1 (ω0−ω1−ω2)p −(ω0+ω1+ω2)p
16N κ ,ν κ ,ν
1 1 2 2
ih
io
h
+ f1Q−f2Q (ω0+ω1−ω2)−p1 − (ω0−ω1+ω2)−p1
N,3n 3n
N,3n
~ X X −κκ0 κ 0 0 −κκ1 κ 1 0 h Q i −1
~ X κ 0 κ 1 −κκ0 −κκ1 h Q i
−
Φ( ν0 ν1 ν2)Φ( ν0 ν1 ν2) 2f1 +1 (ω2)p +
Φ(ν0 ν1 ν0 ν1) 2f1 +1 (7)
8N κ ,ν ν
8N κ ,ν
1
1
2
1
1
and
ΓQ
0 =
N,3n N,3n
π~ X X κ 0 κ 1 κ 2 2nh Q Q i
|Φ(ν0 ν1 ν2)| f1 +f2 +1 δ(ω0−ω1−ω2)
16N κ ,ν κ ,ν
1 1 2 2
h
i
o
+ f1Q−f2Q [δ(ω0+ω1−ω2) − δ(ω0−ω1+ω2)] , (8)
where ( )p and δ( ) denote the Cauchy principal value and Dirac delta function, and ω1
¡
¢
··· κ j
implies ω(κν11). The term Φ κν00 κν11 ···
contains the j th -order derivative of the potential energy
νj
evaluated at the average atomic positions (i.e. the j th -order force constant) and is
¯
3,n 3,n,N
3,n,N
¯
X
X
¡κ 0 κ 1 ··· κ j ¢ X
∂j Φ
¯
¡ lj ¢ ¯
¡ l1 ¢
¡0¢
Φ ν0 ν1 ··· νj =
···
δ(κκ0 +κκ1 +···+κκj ),K
∂rα0 b0 ∂rα1 b1 ··· ∂rαj bj ¯
α0 ,b0 α1 ,b1 ,l1 αj ,bj ,lj
o
(9)
¡ b0 ¢ ¡ b1 ¢
¡ bj ¢
h
h
e0 α0 e1 α1 ··· ej αj
¡ 0 ¢i h
¡ l1 ¢i
¡ lj ¢i
κ0 · r b0 exp iκ
κ1 · r b1 ··· exp iκ
κj · r bj ,
×£
¤ 1 exp iκ
mb0 ω0 mb1 ω1 ··· mbj ωj 2
¡ ¢
where ej αbjj is the eigenvector component associated with atom bj and the αj -direction. The
Kronecker delta, δ(κκ1 +κκ2 +···+κκi ),K , is one if the sum of the wave vectors is a reciprocal lattice
vector, K, and zero otherwise. We use equations (7) and (8) to evaluate the frequency shifts
4
TABLE I: Quantum and classical expressions for phonon properties. In these expressions, V is the
system volume, x ≡ ~ωA /kB T , ωA = ω + ∆, and κx is the wave vector along the x-direction, which
we choose to be [100].
Property
Quantum
Classical
f
1
exp(x)−1
1
x
cph
kB x2 exp(x)
V [exp(x)−1]2
kB
V
vg,x
∂ωA
∂κx
∂ωA
∂κx
τ
1
2Γ
1
2Γ
and linewidths for a classical system by substituting f C −1/2 in place of f Q . The expressions
for the phonon occupation number for quantum and classical harmonic oscillators are given
in Table I.
The thermal conductivity is then evaluated using the relation
k=
XX
κ
2 κ
cph(κν) vg,x
(ν) τ(κν) ,
(10)
ν
which is derived from the Boltzmann transport equation under the relaxation time approximation and is valid for both the quantum and classical systems.23 Expressions for the phonon
specific heat, cph , x-component of the group velocity, vg,x , and relaxation time, τ , are given
in Table I.
We will use the LD method to predict the thermal conductivity of isotopically-pure,
crystalline silicon modeled with the Stillinger-Weber (SW) interatomic potential24 between
temperatures of 10 and 1000 K. While the low-temperature approximations inherent in the
LD techniques cause them to loose accuracy at high temperature, the quantum and classical
thermal conductivity predictions remain self-consistent.23 The thermal conductivity predictions only account for phonon-phonon scattering and neglect other scattering mechanisms,
such as impurity and boundary scattering, that would be present in a real sample. We
perform the calculations on the eight-atom unit cell arrayed on a simple cubic N0 × N0 × N0
lattice. We set the lattice parameter to 5.43 Å, which is within 0.4% of the MD-predicted
lattice parameters at all temperatures considered.25 To achieve size-independent thermal
conductivity predictions, we use N0 = 12 for temperatures above 100 K and N0 = 30 for
lower temperatures. Further details about LD theory and our implementation of the LD
5
100000
Debye
Temperature
k (W/m-K)
10000
1000
Quantum
Classical
QCs w/ ZP
QCs w/o ZP
100
10
100
1000
T (K)
FIG. 1: (Color online) Thermal conductivity predictions for the quantum and classical systems
and the quantum-corrected classical prediction using the ZP energy (QCs w/ ZP) and neglecting
it (QCs w/o ZP).
method for predicting thermal conductivity are given by Turney et al.23
III.
A.
ASSESSING THE QUANTUM CORRECTIONS
Quantum and classical thermal conductivity
The SW silicon thermal conductivities predicted using the LD method are plotted in
Fig. 1 for the quantum and classical systems between temperatures of 10 and 1000 K.
The quantum and classical predictions converge at high temperature, as they must. The
low-temperature thermal conductivity behavior is also as expected. The quantum thermal
conductivity peaks at a temperature of 30 K then approaches zero as the temperature goes
to zero. The classical thermal conductivity prediction increases monotonically as the temperature decreases and diverges to infinity at zero-temperature. For temperatures less than
100 K, phonon scattering from defects, isotopes, and boundaries is important.26 We have
neglected these scattering mechanisms, resulting in very large thermal conductivity values
at these temperatures. We note that Broido et al.26 predicted the thermal conductivity of
SW silicon using an approach related to the LD method presented here with these additional
scattering mechanisms included.
For temperatures above 15 K, the classical thermal conductivities are less than the quan-
6
tum values. This result is not necessarily intuitive because the classical phonon specific heats
are always greater than their quantum counterparts and, according to Eq. (10), will tend to
increase the classical thermal conductivity over the quantum value. The thermal conductivity, however, is also affected by the group velocities and the relaxation times. Because
SW silicon is a stiff material, the group velocities for the quantum and classical systems are
nearly identical and show weak temperature dependence. The relaxation times, however,
have a large impact.
Plotted in Fig. 2 are the phonon linewidths ( 2τ1 ) versus harmonic frequency at a temperature of 100 K. At frequencies below 80 rad/ps, the quantum phonons have smaller linewidths
than the classical phonons. At higher frequencies, the situation is reversed. Considering the
equation for the linewidth [Eq. (8)], we see that two types of phonon interactions contribute
to Γ0 . One is the decay of the phonon ω0 into two phonons (ω0 → ω1 + ω2 , a type I interaction). The other is the annihilation of the phonon with a second to create a third
(ω0 + ω1 → ω2 or ω0 + ω2 → ω1 , a type II interaction). The differences between the quantum
and classical linewidths are driven by the occupation numbers, f Q and f C , which weight
the phonon interactions. For the quantum and classical systems, the type I interactions are
weighted by f1Q + f2Q + 1 and f1C + f2C . Of these two weighting factors, the quantum expression is always greater than the classical expression for the same set of frequencies at the
same temperature. The type II weighting factor for the quantum system, f1Q − f2Q , is always
less than its classical counterpart, f1C − f2C . We note that energy conservation, enforced by
the Dirac delta functions in Eq. (8), causes type I interactions to dominate when ω0 is large,
resulting in quantum linewidths that are greater than their classical counterparts. When
ω0 is small, type II interactions dominate, giving rise to quantum linewidths that tend to
be smaller than the classical values. At intermediate frequencies, both type I and type II
interactions play an important role. This discussion illustrates an important point, which
is that the relaxation time for a single phonon mode depends strongly on the other phonon
modes in the system.
To explore the consequences of the different relaxation times on the thermal conductivQ
C
ity, consider Fig. 3 where the ratio of the mode-dependent thermal conductivities, kph
/kph
Q
C
C 2 C
Q 2 Q C
[= cQ
ph (vg,x ) τ /cph (vg,x ) τ ], is plotted against the ratio of the relaxation times, τ /τ ,
Q
C
at a temperature of 100 K. We identify four regions separated by the lines kph
/kph
= 1,
Q
C
τ Q /τ C = 1, and kph
/kph
= τ Q /τ C . Region IV is inaccessible because the quantum specific
7
0.08
T=100 K
Γ (rad/ps)
0.06
Quantum
0.04
Classical
0.02
0
0
20
40
FIG. 2: (Color online) Phonon linewidths (Γ =
60
ω (rad/ps)
1
2τ )
80
100
120
plotted versus harmonic frequency for quantum
and classical systems at a temperature of 100 K.
100
T=100 K
10
III
1
Q
kph /kph
C
IV
0.1
I
II
0.01
0.001
0.1
1
10
100
τQ/τC
FIG. 3: Ratio of the quantum to classical phonon thermal conductivities plotted versus the relaxation time ratio for all phonon modes at a temperature of 100 K.
heat is always lower than the classical specific heat. In regions I and II, the mode-dependent
quantum thermal conductivity is lower than the classical value, despite the quantum relaxation time being larger than the classical relaxation time in region II. For phonon modes
in region III, the quantum relaxation time is so much larger than the classical value that
it negates the reduction caused by the specific heat. The net result in this region is that
the mode-dependent quantum thermal conductivity is larger than the classical value. For
temperatures above 15 K, the phonons in region III dominate the thermal transport, causing
8
the total quantum thermal conductivity to be greater than the classical prediction (see Fig.
1).
B.
Quantum-corrected thermal conductivity
The temperature mapping from the classical system to the quantum system defined by
Eq. (2) is plotted in Fig. 4 when the ZP energy is both considered and neglected. Also
plotted is the scaling factor dT C /dT Q defined by Eq. (4). This factor is independent of the
ZP energy and is always less than one. The scaling factor can be written as
N,3n
X Q
dT C
1
=
cph(κν) ,
Q
dT
3(N n − 1)kB κ ,ν
(11)
which is obtained by differentiating Eq. (2) with respect to T Q and assuming the phonon
frequencies to be temperature-independent. The temperature mapping depends strongly on
the inclusion or exclusion of the ZP energy. For the same quantum temperature, the classical
temperatures for the two cases are offset by the average ZP energy divided by kB , which is
270 K for SW silicon. When the ZP energy is included, the corrected temperature is greater
than the quantum temperature, the corrected and quantum temperatures converge in the
high-temperature (i.e. classical) limit, and a classical system below a temperature of 270 K
has no quantum counterpart. When the ZP energy is excluded, the corrected temperature
is less than the quantum temperature and the corrected and quantum temperatures do not
converge in the high-temperature limit, but are separated by 270 K.
The effect of the QCs on the classical thermal conductivity prediction is shown in Fig.
1, where we plot the quantum-corrected classical predictions. Neglecting the ZP energy
increases the thermal conductivity. Including the ZP energy causes the thermal conductivity
to shift downward, away from the quantum predictions.
For QCs to be valid, the corrected classical predictions must (i) converge to the quantum
predictions at high temperature and (ii) provide a better estimation of the quantum thermal
conductivity than the uncorrected classical predictions. Our four sets of results (quantum,
classical, and two sets of quantum-corrected classical values) do converge at high temperature, though the convergence is slow for both sets of quantum-corrected values. At the
SW silicon Debye temperature of 710 K,27 the corrected thermal conductivities with and
without the ZP energy are 10% lower and 30% higher than the quantum value, while the
9
1000
1.0
C
dT
dT Q
0.8
)
ZP
/
sw
600
T
400
/o
sw
T
200
0.6
C
(Q
(Q
)
ZP
C
0.2
0
0
200
400
0.4
dT C/dT Q
T C (K)
800
600
800
0
1000
T Q (K)
FIG. 4: (Color online) Mapping between quantum and classical temperatures with and without
the ZP energy and dT C /dT Q as defined by Eqs. (2) and (4).
uncorrected classical prediction is within 5% of the quantum value. At a temperature of
300 K, less than half the Debye temperature, the classical prediction is still within 15% of
the quantum value. For temperatures above 200 K, neither of the QC approaches improves
upon the agreement between the uncorrected classical value and the quantum value.
When including the ZP energy, a maximum in the thermal conductivity is predicted.
The location and magnitude of this maximum, however, are clearly wrong. The lowest
uncorrected temperature we consider is 10 K, which corresponds to a temperature of 105 K
for the QCs without the ZP energy. The quantum-corrected thermal conductivity predictions
without the ZP energy can be extended to lower temperatures by fitting a power function
to the uncorrected classical predictions and extrapolating. In doing so (not shown), we find
that the quantum-corrected thermal conductivity does not exhibit a maximum when the
ZP energy is neglected. From these observations, we conclude that the QCs prescribed by
Eqs. (2) and (4), with the ZP energy either included or neglected, do not properly account
for quantum effects. These QCs even fail at high temperatures, where quantum effects are
small.
10
k contribution
(arb. units, log scale)
10 K
100 K
500 K
Classical
0
0.2
0.4
0.6
ωA/(ωA)max
0.8
1
FIG. 5: (Color online) Scaled contribution to the thermal conductivity as a function of frequency
ratio. The classical results at all temperatures collapse to a single curve.
C.
Analyzing the quantum corrections
The reasons why the QCs given by Eqs. (2) and (4) fail can be deduced by considering
the mode-dependence of the phonon properties (ω, cph , vg,x , and τ ). In Fig. 5, we plot the
contribution to the thermal conductivity versus phonon frequency, both scaled by their maximum values, for temperatures of 10, 100, and 500 K. When plotted this way, the results for
the classical systems collapse to a single, temperature-independent curve. For the quantum
system, however, the low-frequency phonons increasingly dominate the thermal conductivity as the temperature decreases. This effect is due to the temperature-dependence of the
occupation number [Eq. (3)], which effectively freezes out the high-frequency modes at low
temperatures. Applying the QCs given by Eqs. (2) and (4) shifts the thermal conductivity by
a temperature-dependent scale factor. Yet, we see from Fig. 5 that the frequency-dependence
of the mode contribution is strongly temperature-dependent. The thermal conductivity thus
cannot be properly corrected by applying a system-level scaling factor.
We can further address the reasons why the QCs fail by investigating what would need to
be done to properly correct a classically-predicted thermal conductivity. From the discussion
in the previous paragraph, we know that QCs must be performed on the level of the phonon
modes [i.e., the terms inside the summation in Eq. (10)]. To simplify the analysis, we
retain the assumption of a temperature-independent lattice constant and further assume the
frequency shift to be negligible. This assumption is good for SW silicon as the root mean
11
1000
T C = 1000 K
T Q (K)
800
T C = 710 K
600
400
T C = 420 K
200
T C = 100 K
0
0
20
40
60
ω (rad/ps)
80
100
FIG. 6: Frequency-dependent quantum temperatures as determined by Eq. (12) for various classical
temperatures. For temperatures less than 420 K, some of the high-frequency phonons cannot be
mapped into any quantum system.
square of ∆/ω is less than 0.03 for all temperatures considered. When the frequency shift is
neglected, the x-component of the group velocity becomes temperature-independent and is
the same in the quantum and classical systems. The phonon specific heats and relaxation
times, however, differ between the two systems. If these two mode-dependent properties
can be properly quantum-corrected, then we would be able to predict a quantum thermal
conductivity from classical phonon properties. In what follows, we will assume that all the
classical phonon properties are known. Though not typically found in a Green-Kubo28,29
or direct28,30 MD-based thermal conductivity prediction, these phonon properties can be
obtained from MD simulations.23,31
Correcting the specific heat is straight-forward. The classical specific heat is a constant
and the quantum specific heat can be computed from the phonon frequency (see Table
I). The first step in converting the classical thermal conductivity to a quantum thermal
C 32
conductivity thus involves scaling the contribution of each phonon mode by cQ
From
ph /cph .
Eq. (11), we see that the quantum correction defined by Eq. (4) is on the right track in that
it scales the classical thermal conductivity by the ratio of the quantum specific heat to the
classical specific heat. This scaling, however, is performed on the system level, rather than
on the level of the phonon modes, as is required.
The relaxation times cannot be corrected in the same manner. Using the LD approach
described in Section II, the effort required to compute quantum relaxation times is the same
12
as the effort required to compute classical relaxation times. Additionally, if the quantum
relaxation times were known, we could compute the quantum thermal conductivity directly
using Eq. (10).
What we would like is for the classical relaxation times to be representative of the relaxation times of a corresponding quantum system. We determine the quantum system
represented by the classical relaxation times by setting τ Q = τ C for each phonon mode,
which is equivalent to letting ΓQ = ΓC . The equation for the quantum linewidth [Eq. (8)]
can be converted to a classical linewidth by setting
f C = f Q + 1/2.
(12)
This condition links the quantum and classical systems. Solving Eq. (12) for T Q , we find
the temperature of a quantum system that has the same occupation number as the classical
system to be
TQ =
h
kB ln
~ω
2kB T C +~ω
2kB T C −~ω
i.
(13)
We can immediately see a problem with Eq. (13). The quantum temperature is a function
of the phonon frequency, meaning that the classical relaxation times cannot be mapped
onto an equilibrium quantum system at a single temperature. The quantum temperature
corresponding to classical temperatures of 100, 420, 710, and 1000 K is plotted versus frequency in Fig. 6. Above the Debye temperature (710 K), the range of quantum temperatures
associated with the classical temperature is small. The classical relaxation times at these
temperatures are a good approximation of the quantum relaxation times, and there is no
need to perform a quantum correction. As the classical temperature decreases below the Debye temperature, the range of quantum temperatures increases and the classical relaxation
times are not indicative of the relaxation times of an equilibrium quantum system. When the
classical temperature drops below 420 K, some of the high-frequency phonon modes cannot
be mapped onto any quantum system. Though they make a small direct contribution to
the thermal conductivity, these high-frequency phonons are important in that they scatter
with low-frequency phonons. These mapping issues are also inherent in Eq. (2) (where T Q
is found as a weighted average) and contribute to the failure of the QCs.
13
D.
Discussion
Others have argued for the QCs given by Eqs. (2) and (4) because they can bring classical MD thermal conductivity predictions into better agreement with experimental measurements. When examining all the available data, however, we find that some find better
agreement when the ZP energy is considered13,14,16,32 while others find better agreement
when the ZP energy is ignored.11,12 Based on the results presented in Section III B, we
believe that any improved agreement is fortuitous as MD-predicted thermal conductivities
are sensitive to the chosen interatomic potential. Interatomic potentials are an approximation of the complex interactions that exist between atoms, which only quantum calculations
(e.g., density functional theory) can accurately capture. Using an approach related to the
LD method presented here, Broido and co-workers found that a fully quantum-mechanical
treatment of SW silicon does not match the experimental thermal conductivity data,26 while
the same treatment using input from density functional theory calculations does.33
Alternative approaches for obtaining a quantum thermal conductivity prediction from
classical MD simulations have been suggested. These approaches are based on modifying
the MD simulation itself, making them fundamentally different than QCs, which are a postprocessing procedure. Li suggested initializing a classical MD simulation with a phonon
distribution corresponding to a quantum system at equilibrium.34 Wang proposed a classical
MD simulation that interacts with heat baths that are forced to maintain quantum phonon
distributions.35 The drawback of these approaches is that the MD simulation is limited
temporally or spatially. A system with an initial quantum configuration will eventually
relax to an equilibrium classical state. A large MD system in contact with quantum heat
baths will have a classical phonon distribution far from the heat baths.
Quantum MD simulation (e.g., Car-Parrinello36 ) in conjunction with the Green-Kubo or
direct method has not been used to predict thermal conductivity because it is too computationally expensive. A less computationally expensive approach is to use Feynman path
integrals to effectively transform a quantum system into a classical system, whose static
properties can then be computed using classical MD or Monte Carlo simulations.37 Centroid
MD, an extension of path-integral MD developed by Cao and Voth,38,39 has been used to
predict transport properties of quantum systems, such as thermal conductivity and shear
viscosity.40
14
Fully quantum-mechanical predictions of thermal conductivity, made using the LD
method employed here or a related method developed by Omini and Sparavigna,41,42 are
an excellent alternative to MD-based predictions for ordered systems well below the Debye
temperature, where the weakly-interacting phonon model is valid. Lattice dynamics-based
predictions of the thermal conductivity have the additional advantages of allowing their input to come from density functional theory calculations and providing the complete phonon
properties.23 At higher temperatures, quantum effects are unimportant and classical MD
based methods are suitable if an accurate interatomic potential is availible.
IV.
CONCLUSIONS
We have assessed the validity of the commonly-used QCs for thermal conductivity [Eqs.
(2) and (4)] by using harmonic and anharmonic LD calculations to self-consistently predict
the quantum and classical thermal conductivities of SW silicon. Applying the QCs to
the classical predictions, with or without the ZP energy, does not bring them into better
agreement with the quantum predictions compared to the uncorrected classical values above
temperatures of 200 K (see Fig. 1). When neglecting the ZP energy, the quantum-corrected
temperature does not approach the quantum temperature in the high-temperature limit
and there is no maximum in the thermal conductivity. When the ZP energy is included, the
corrected thermal conductivity shifts away from the quantum prediction and the location and
magnitude of the maximum do not agree with the quantum results. Though not presented
here, we have seen the same behavior in Lennard-Jones argon.
By examining the frequency dependence of the thermal conductivity (see Fig. 5), we
found that system-level properties, like those used in Eqs. (2) and (4), cannot be used
to bring classical thermal conductivities into agreement with quantum values. While the
specific heat can be quantum-corrected on a mode-by-mode basis, the system temperature
cannot be quantum-corrected because the phonons in a classical system are not representative of the phonons in an equilibrium quantum system at a single temperature (see Fig. 6).
Differences between the classical and quantum relaxation times (see Figs. 2 and 3) can be
attributed to the frequency dependence of the occupation numbers and the dominant scattering mechanism. The mapping of classical phonons onto an equivalent quantum system is
approximately correct only at high temperatures, where QCs are unnecessary.
15
Acknowledgments
This work is supported in part by the Pennsylvania Infrastructure Technology Alliance, a
partnership of Carnegie Mellon, Lehigh University, and the Commonwealth of Pennsylvania’s
Department of Community and Economic Development (DCED). Additional support is
provided by Advanced Micro Devices (AMD), and the Berkman Faculty Development Fund
at Carnegie Mellon University. We also thank E. S. Landry (Carnegie Mellon University)
for providing suggestions to improve the manuscript.
∗
Corresponding author: [email protected]
1
M. P. Allen and D. J. Tildesley, Computer Simulation of Liquids (Oxford University Press,
Oxford, 1987).
2
D. C. Rapaport, The Art of Molecular Dynamics Simulation (Cambridge University Press,
Cambridge, 1995).
3
A. J. H. McGaughey and M. Kaviany, in Advances in Heat Transfer, edited by G. Greene,
Y. Cho, J. Hartnett, and A. Bar-Cohen (Elsevier, 2006), vol. 39, pp. 169–255.
4
P. H. Berens, D. H. J. Mackay, G. M. White, and K. R. Wilson, The Journal of Chemical
Physics 79, 2375 (1983).
5
M. Matsui, The Journal of Chemical Physics 91, 489 (1989).
6
V. A. Levashov, S. J. L. Billinge, and M. F. Thorpe, Journal of Computational Chemistry 28,
1865 (2007).
7
J. Che, T. Cagin, W. Deng, and W. A. Goddard III, The Journal of Chemical Physics 113,
6888 (2000).
8
C. Z. Wang, C. T. Chan, and K. M. Ho, Physical Review B 42, 11276 (1990).
9
Y. H. Lee, R. Biswas, C. M. Soukoulis, C. Z. Wang, C. T. Chan, and K. M. Ho, Physical Review
B 43, 6573 (1991).
10
A. Maiti, G. D. Mahan, and S. T. Pantelides, Solid State Communications 102, 517 (1997).
11
S. G. Volz and G. Chen, Physical Review B 61, 2651 (2000).
12
S. Volz and G. Chen, Physica B 263-264, 709 (1999).
13
C. J. Gomes, M. Madrid, J. V. Goicochea, and C. H. Amon, Journal of Heat Transfer 128, 1114
16
(2006).
14
J. Li, L. Porter, and S. Yip, Journal of Nuclear Materials 255, 139 (1998).
15
S. Volz, J. B. Saulnier, G. Chen, and P. Beauchamp, Microelectronics Journal 31, 815 (2000).
16
J. R. Lukes and H. Zhong, Journal of Heat Transfer 129, 705 (2007).
17
P. Heino, Physica Scripta T114, 171 (2004).
18
A. J. H. McGaughey, Ph.D. thesis, University of Michigan, Ann Arbor, Michigan (2004).
19
M. T. Dove, Introduction to Lattice Dynamics (Cambridge University Press, Cambridge, UK,
1993).
20
A. J. H. McGaughey, M. I. Hussein, E. S. Landry, M. Kaviany, and G. M. Hulbert, Physical
Review B 74, 104304 (2006).
21
D. C. Wallace, Thermodynamics of Crystals (John Wiley & Sons, New York, 1972).
22
A. A. Maradudin and A. E. Fein, Physical Review 128, 2589 (1962).
23
J. E. Turney, E. S. Landry, A. J. H. McGaughey, and C. H. Amon, Physical Review B 79,
064301 (2009).
24
F. Stillinger and T. Weber, Physical Review B 31, 5262 (1985).
25
J. Goicochea, M. Madrid, and C. Amon, Journal of Heat Transfer (in press, 2009).
26
D. A. Broido, A. Ward, and N. Mingo, Physical Review B 72, 014308 (2005).
27
E. S. Landry and A. J. H. McGaughey, Physical Review B 79, 075316 (2009).
28
P. K. Schelling, S. R. Phillpot, and P. Keblinski, Physical Review B 65, 144306 (2002).
29
H. Kaburaki, J. Li, S. Yip, and H. Kimizuka, Journal of Applied Physics 102, 043514 (2007).
30
J. Lukes, D. Li, X.-G. Liang, and C.-L. Tien, Journal of Heat Transfer 122, 536 (2000).
31
A. J. H. McGaughey and M. Kaviany, Physical Review B 69, 094303 (2004).
32
J. V. Goicochea, Ph.D. thesis, Carnegie Mellon University, Pittsburgh, PA (2008).
33
D. A. Broido, M. Malorny, G. Birner, N. Mingo, and D. A. Stewart, Applied Physics Letters
91, 231922 (2007).
34
J. Li, Ph.D. thesis, Massachusetts Institute of Technology, Cambridge, MA (2000).
35
J.-S. Wang, Physical Review Letters 99, 160601 (2007).
36
R. Car and M. Parrinello, Physical Review Letters 55, 2471 (1985).
37
D. M. Ceperley, Reviews of Modern Physics 67, 279 (1995).
38
J. Cao and G. A. Voth, The Journal of Chemical Physics 100, 5093 (1994).
39
J. Cao and G. A. Voth, The Journal of Chemical Physics 100, 5106 (1994).
17
40
Y. Yonetani and K. Kinugawa, The Journal of Chemical Physics 119, 9651 (2003).
41
M. Omini and A. Sparavigna, Physica B 212, 101 (1995).
42
M. Omini and A. Sparavigna, Physical Review B 53, 9064 (1996).
18